Code Red

Good morning.

Interesting developments this week. Drama at OpenAI, once again. Sam Altman has called a “Code Red”. Gemini, Claude and DeepSeek have all surpassed GPT-5 in capability over the last few weeks. And while this is not new, this time we see people actually abandoning ChatGPT for particularly Gemini. ChatGPT’s subscription base is dropping. We don’t know how much, because we don’t have those numbers but it must be ugly.

You could also feel the lack of excitement on their latest “Dev Days” - updates are lacklustre.

I must admit, I am one of the people that switched. The reason for that is simple : Gemini feels smarter and sharper. But I think the real reason is not because these models are more powerful (they are) - it’s because they don’t restrict you as much.

With ChatGPT there are constraints and limits - for example if you dive into something medical you will be warned multiple times you should go to a real doctor. The other models don’t paternise you and they treat you as an adult.

The real problem is that ChatGPT has constraints and ideological biases. The model is politically on the left. Which feels very strange in certain interactions. People don’t like being told what to do or what to think.

It feels similar to when Disney released the new Snow White movie earlier this year. This movie went overboard with “woke” ideology. For example the Prince couldn’t kiss Snow White awake because she couldn’t give consent. Things like that.

This is now in contention for being the worst movie of all time on IMDB (2.2). This is the way people are letting Disney know they don’t like what is happening.

It is well known that Silicon Valley - just like Hollywood - is its own bubble and is prone to adopting this ideology and taking it too far. Equality and diversity are important but this movie shows you can take it too far. I think same goes for OpenAI and ChatGPT. When you tell someone how to think even in a very subtle way, they rebel.

It’s not only political bias, if you want to debate for example the presence of non-human intelligence on earth then GPT-5 is unwilling to have an open mind about that, for example. I read on X that someone tried this and so I had to do it myself. So I tried pushing it and at one point it even called me a ‘conspiracy theorist’ and that I shouldn’t entertain these thoughts. While Gemini or Claude are gladly coming up with theories.

It will be very interesting to see how OpenAI are going to pivot out of this.

When talking to “the man in the street” (or the woman). (See how I’m already affected by the woke mind virus :)) - Talking to the common man - “ChatGPT” is synonymous with AI.

That might change in a few months.

Welcome to the Blacklynx Brief

AI News

• DeepSeek has released an open-source AI model, DeepSeek-Math-V2, that can solve advanced mathematical problems at the level of a major international competition by using a built-in system that forces the model to debug and correct its own reasoning.

• OpenAI reported that a security breach at its analytics partner, Mixpanel, exposed the basic profile details—like names and emails—of some API users, though no sensitive data such as chat history or passwords were affected.

• New research from NVIDIA and a Hong Kong university shows that training a small AI to act as a "conductor" for specialized tools can outperform much larger AIs more efficiently, suggesting that smart system coordination may be the next big breakthrough.

• The AI system named Aristotle, created by the company Harmonic, recently solved a version of a long-unanswered math problem posed 30 years ago by Paul Erdős, achieving the feat in hours and then verifying the complex proof in a minute. This breakthrough signals the start of what researchers call the "vibe proving" era, where AI rapidly discovers proofs that are then rigorously verified by machine tools. This development further pushes the frontier of mathematical superintelligence, potentially allowing more people to contribute to advanced mathematics.

• A study by MIT and Hugging Face revealed a major shift in the global open-source AI landscape, showing that Chinese developers have surpassed the U.S. industry in terms of model downloads for the first time. This rise is mostly driven by two Chinese companies, DeepSeek and Alibaba, which together hold a large market share, indicating a significant change in who leads the development of public AI technologies. The report also noted a trend toward less transparency, as fewer new models are disclosing the data used to train them.

• The Chinese AI firm DeepSeek has released two new powerful reasoning models, V3.2 and V3.2-Special, which match or even surpass the performance of top-tier proprietary systems like GPT-5 and Gemini 3 Pro on tests like advanced math and coding. The company has made these high-performing models openly available to the public under an MIT license, but most notably, their operating cost is a small fraction of the price charged by U.S. competitors. This release puts significant pressure on major labs to justify their high premium fees for access to frontier AI technology.

• The AI video company Runway has launched its latest model, Gen-4.5, which claims to achieve a new level of realism and creative control in text-to-video generation, now ranking at the top of independent industry leaderboards. Runway states that the new model handles complex details like physics, fabric, and human motion with greater naturalness and is capable of generating cinematic visuals that are difficult to distinguish from real footage. This rapid year-over-year improvement brings AI video much closer to the high cinematic quality needed for widespread professional adoption in industries like Hollywood.

• The Chinese startup Kuaishou has launched Kling O1, a new comprehensive AI system designed to handle both the creation and complex editing of videos within a single, unified tool. Users can generate short clips and also modify existing footage with simple text commands, allowing for granular edits such as removing objects or changing the time of day while keeping the main characters consistent. This all-in-one capability marks a significant step in making detailed and creative video editing much more accessible and efficient than ever before.

• Facing intensified competition from rivals like Google, OpenAI CEO Sam Altman has reportedly declared an internal "code red" to immediately shift priorities and accelerate major improvements to the ChatGPT user experience, particularly focusing on personalization and image generation. The company is fast-tracking the launch of a new reasoning model said to outperform Google's latest offering, while also preparing for a larger 2026 upgrade codenamed "Garlic" to ensure it maintains its technological lead. This move signals that the competition for dominance in the consumer AI market is becoming extremely intense, forcing established leaders to react quickly.

• At its annual conference, Amazon Web Services (AWS) unveiled a major set of new AI offerings, including an upgraded family of powerful models called Nova 2, a service that lets businesses train their own custom versions of these models, and a more advanced AI chip, Trainium 3. AWS also heavily invested in AI agents by launching three specialized "frontier agents" for tasks like coding and security, which can run autonomously for long periods, showing a strong push to compete across the entire spectrum of AI technology from hardware to enterprise tools.

• The French AI startup Mistral has released Mistral 3, a large family of ten new AI models with open-source weights, which includes its flagship Large 3 model and nine smaller, more efficient versions called Ministral. The smaller Ministral models are designed to run on common devices like laptops and phones without needing a continuous internet connection, which opens up new opportunities for using advanced AI in a wide variety of personal and on-device applications. While the main model competes well with others, the most significant impact may come from the smaller models that bring sophisticated AI to the edge.

Quickfire News

• OpenAI lost a key discovery ruling: OpenAI must hand over internal communications regarding why it deleted two datasets of allegedly pirated books, which helps authors in their copyright infringement case.

• Alibaba released Quark AI Glasses: Alibaba launched a line of smart eyewear in China, powered by its in-house Qwen AI models and assistant, starting at 1,899 yuan ($268).

• Perplexity updated its email assistant: Perplexity's email assistant can now manage multiple calendars simultaneously and is available for both Gmail and Outlook users.

• Jeff Bezos’ new stealth AI venture acquired General Agents: Jeff Bezos' secret AI company, "Project Prometheus," quietly purchased General Agents, a startup focused on advanced agentic computing, according to Wired.

• Cohere expanded its partnership with SAP: Cohere's agentic AI platform, North, will now be integrated into SAP’s Cloud infrastructure and Business Technology Platform.

• Perplexity launched persistent memory: Perplexity introduced a feature that lets its assistant remember a user's preferences, interests, and past conversations for better context on relevant tasks.

• Databricks is in talks to raise funds: The data intelligence company Databricks is reportedly discussing a plan to raise $5 billion, which would set its total value at $134 billion, according to The Information.

• Deutsche Telekom and Schwarz Group plan "AI gigafactory": These two companies are reportedly collaborating to build a massive "AI gigafactory" in Germany and may seek part of the EU's $20 billion funding to compete with U.S. and Chinese tech rivals.

• Telegram CEO launched a decentralized compute network: Pavel Durov of Telegram introduced Cocoon, a network allowing people with powerful GPUs (graphics processing units) to earn TON tokens by providing their hardware for private AI processing, competing with traditional cloud services.

• Elon Musk’s xAI is set to raise significant funding: Elon Musk’s AI company, xAI, is reportedly planning to raise $15 billion next month, which would value the company at $230 billion before the new funds, as reported by CNBC.

• AI and agents drove major Black Friday sales: Data from Salesforce shows that AI and automated shopping agents were responsible for $14.2 billion in online sales globally on Black Friday, with $3 billion of that total coming from the U.S.

• Avatar director James Cameron criticized generative AI: Director James Cameron called generative AI "horrifying," arguing it creates a performance from nothing, unlike his performance capture technique which focuses on the real "actor-director moment."

• Virgin Australia signed a deal with OpenAI: Virgin Australia partnered with OpenAI to embed ChatGPT-powered tools into its flight search and planning process for customers.

• Black Forest Labs announced a new funding round: Black Forest Labs secured a new $300 million investment, valuing the company at $3.25 billion, following the launch of their Flux.2 image model.

• Epic CEO Tim Sweeney is lobbying against ‘Made with AI’ tags: The CEO of Epic is pushing for game marketplaces like Steam to stop using 'Made with AI' labels, arguing that AI will soon be involved in almost all production.

• Accenture and OpenAI are partnering: Accenture and OpenAI are working together to give tens of thousands of Accenture consultants access to ChatGPT Enterprise and to help clients use AI agents in their businesses.

• OpenAI is taking an ownership stake in Thrive Holdings: OpenAI is acquiring a stake in Thrive Holdings, a company owned by one of its investors, Thrive Capital, in a deal intended to scale its impact on enterprise operations.

• Nvidia invested $2 billion in Synopsys: Nvidia invested $2 billion in the chip design software company Synopsys as part of a multi-year partnership focused on using AI and computing to speed up product engineering.

• Apple announced a leadership change in AI: Apple's current AI chief, John Giannandrea, will retire in early 2026, and former Microsoft and Google executive Amar Subramanya has been hired to take over major AI responsibilities.

• NVIDIA unveiled an AI model for self-driving cars: NVIDIA introduced Alpamayo-R1, an open model designed for self-driving vehicles, which uses step-by-step reasoning to handle complex situations like intersections crowded with pedestrians.

• OpenAGI emerged with a new AI model: The company OpenAGI came out of its secretive development phase with "Lux," an AI model created to control computers independently, and which reportedly performs better than similar models from Google, OpenAI, and Anthropic.

• Microsoft CEO discussed AI's energy demands: Satya Nadella, CEO of Microsoft, stated that the AI industry must "earn the social permission" from the public to justify the massive amount of energy it requires, which is putting a strain on electrical grids.

• Anthropic acquired Bun and shared revenue news: Anthropic purchased the open-source JavaScript toolkit known as Bun, and also announced that their Claude Code AI product has reached $1 billion in annual revenue just six months after its release.

Closing Thoughts

That’s it for us this week.